June 2022
Volume 63, Issue 7
Open Access
ARVO Annual Meeting Abstract  |   June 2022
Validation of an OCT-based deep-learning algorithm for the identification of full-thickness idiopathic macular holes (FTIMH)
Author Affiliations & Notes
  • Carolina Carvalho Soares Valentim
    Ophthalmology, Center for Ophthalmic Bioinformatics, Cleveland Clinic Cole Eye Institute, Cleveland, Ohio, United States
  • Anna K Wu
    Ophthalmology, Center for Ophthalmic Bioinformatics, Cleveland Clinic Cole Eye Institute, Cleveland, Ohio, United States
  • Weilin Song
    Ophthalmology, Center for Ophthalmic Bioinformatics, Cleveland Clinic Cole Eye Institute, Cleveland, Ohio, United States
  • Victoria Wang
    Case Western Reserve University, Cleveland, Ohio, United States
  • Jessica L. Cao
    Ophthalmology, Center for Ophthalmic Bioinformatics, Cleveland Clinic Cole Eye Institute, Cleveland, Ohio, United States
  • Sophia Yu
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Niranchana Manivannan
    Carl Zeiss Meditec, Inc., Dublin, California, United States
  • Rishi P Singh
    Ophthalmology, Center for Ophthalmic Bioinformatics, Cleveland Clinic Cole Eye Institute, Cleveland, Ohio, United States
  • Katherine E Talcott
    Ophthalmology, Center for Ophthalmic Bioinformatics, Cleveland Clinic Cole Eye Institute, Cleveland, Ohio, United States
  • Footnotes
    Commercial Relationships   Carolina Carvalho Soares Valentim None; Anna Wu None; Weilin Song None; Victoria Wang None; Jessica Cao None; Sophia Yu Carl Zeiss Meditec, Inc. , Code E (Employment); Niranchana Manivannan Carl Zeiss Meditec, Inc. , Code E (Employment); Rishi Singh Novartis, Genentech/Roche, Regeneron, Alcon, Zeiss, Bausch and Lomb, Gyroscope, Code C (Consultant/Contractor), Aerie, Apellis, Graybug, Code F (Financial Support); Katherine Talcott Zeiss, Regenxbio, Code F (Financial Support), Genentech/Roche, Code I (Personal Financial Interest)
  • Footnotes
    Support  This study was supported in part by the NIH-NEI P30 Core Grant (IP30EY025585), Unrestricted Grants from The Research to Prevent Blindness, Inc., and Cleveland Eye Bank Foundation awarded to the Cole Eye Institute.
Investigative Ophthalmology & Visual Science June 2022, Vol.63, 2103 – F0092. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      Carolina Carvalho Soares Valentim, Anna K Wu, Weilin Song, Victoria Wang, Jessica L. Cao, Sophia Yu, Niranchana Manivannan, Rishi P Singh, Katherine E Talcott; Validation of an OCT-based deep-learning algorithm for the identification of full-thickness idiopathic macular holes (FTIMH). Invest. Ophthalmol. Vis. Sci. 2022;63(7):2103 – F0092.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : Automated identification of OCT features can improve retina clinic workflow efficiency as they are able to detect pathological findings. The purpose of this study was to validate an OCT-based deep-learning algorithm for the identification of FTIMH features and stages.

Methods : In this retrospective study, subjects solely diagnosed with either FTIMH or Posterior Vitreous Detachment (PVD) were identified excluding secondary causes of macular holes, any concurrent maculopathy, or incomplete records. OCT scans (512x128) from all subjects were acquired with CIRRUSTM HD-OCT 5000 (ZEISS, Dublin, CA) and reviewed for quality. In order to establish a ground truth classification, each OCT B-scan was labeled by two trained graders (Table 1). A retina specialist acted as a tie-breaker whenever there was a labeling disagreement. The FTIMHs were measured using the caliper tool on CIRRUS Review software and classified in stages according to the International Vitreomacular Traction Study. The accuracy of the algorithm to identify disease features in normal and FTIMH OCT B-scans was determined by dividing the number of B-scans in agreement by the total gradable scans labeled by the algorithm. Pearson’s correlation was run to determine if the algorithm’s probability score was associated with the stages of FTIMH.

Results : Ninety-nine OCT cube scans from 99 subjects (49 with FTIMH and 50 with PVD) were used. Among the FTIMH scans, 63% (n=31) were stage 4, 10% (n=5) were stage 3 and 27% (n=13) were stage 2. A total of 12,354 individual OCT B-scans were labeled gradable by the algorithm and yielded an accuracy of 90.6% in identifying OCT features of FTIMHs. A Pearson’s correlation coefficient of 0.31 was achieved between the algorithm’s probability score and the stages of the 49 FTIMHs cubes studied.

Conclusions : The OCT-based deep-learning algorithm was able to accurately detect FTIMHs features on individual OCT B-scans. However, there was a low correlation between the algorithm’s output and FTIMH stages. The algorithm may serve as a clinical decision support tool that assists with the identification of FTIMHs. Further training is necessary for the algorithm to identify stages of FTIMHs.

This abstract was presented at the 2022 ARVO Annual Meeting, held in Denver, CO, May 1-4, 2022, and virtually.

 

Possible labels for OCT B-scans

Possible labels for OCT B-scans

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×